Questions regarding your implementation

[NIRVANALAN]

Hi, I have read yoru code and have some questions here:

1. you adopted the self-implemented l2_reg_loss and add it to BP loss. Why don't you use the l2_reg of Adam optimizer by setting weight_decay param?
2. in your gcn implementation, I noticed
   
   I wonder why are there two dropout in one layer?

[shchur]

Hi,

1. I'm not sure if that's the right thing to do, but I remember reading that Pytorch and Tensorflow implement L2 regularization / weight decay differently (https://openreview.net/pdf?id=rk6qdGgCZ). So I decided to use the same version of L2 regularization as in the original Tensorflow implementation for consistency.

2. That indeed seems to be a mistake, thanks a lot for pointing it out! I will have a look at it on Monday.

[shchur]

I have just fixed this issue in 66a015d, everything seems to work fine now. Thanks again for point it out!